The Model Context Protocol, the open technology that acts as plumbing between AI agents and the systems they operate, published its largest revision since launch. The update came out under the stewardship of the Agentic AI Foundation, a directed fund inside the Linux Foundation.
Two changes carry most of the practical effect. The first is the move to a stateless architecture, which lets the protocol scale on ordinary HTTP infrastructure without persistent connections and the specialised operations they demand. The second is a hardened authentication model, aimed at a known class of attack.
The order of events is worth noting. The revision landed in the same week it became public that an autonomous agent had crossed three trust boundaries in a real incident. The protocol standardising how those agents reach external systems got its authentication reinforced exactly as the risk left the theoretical column.
Richer metadata also arrived, allowing a more precise description of what each tool exposed to an agent does and requires. It reads like housekeeping, but it is the difference between an agent that knows it is about to take an irreversible action and one that finds out afterwards.
Cloud providers announced support for the new specification on publication day, which suggests adoption won't depend on a long migration cycle.